Agent Flow Engine
afe runs agentic flows declared in YAML manifests.
You describe a graph of agents, scripts and human approvals. The engine runs that graph within budgets, loop limits and approval points.
The engine is a standalone component. Everything it talks to is a plugin: model providers, databases, queues, sandboxes, forges, the CLI and the web UI.
The documentation has four sections:
- Tutorials — start here. One guaranteed path from an empty folder to a finished ticket.
- How-to guides — a recipe for one task you already have.
- Reference — every manifest kind and field, the CLI, the API and the operations endpoints.
- Explanation — why the engine is built this way.
New to afe? Start with Your first flow. It uses the dry harness, so it needs no model key.
Read Architecture when you want the reasons behind the design, not just the steps.